Observability

Jaeger

A distributed tracing system: it collects and shows the path of a request through microservices.

Official siteRepository

What it is

Jaeger was created at Uber, open sourced in 2017 and became a CNCF graduated project in 2019. It is an open system for collecting, storing and viewing traces.

It shows how a request travels through services, how long each step takes and where delays or errors happen. The UI has a Gantt chart of spans and a service dependency map.

Today Jaeger works with OpenTelemetry (OTLP) data: apps are instrumented through OTel, and Jaeger acts as the storage backend (Elasticsearch, Cassandra, Badger) and UI.

Use it when

  • A microservice architecture where it is hard to find which service slows a request.
  • You need a self-hosted open source tracing tool.
  • Analyzing latency and dependencies between services.
  • You want a backend for OpenTelemetry data without a commercial vendor.

Avoid it when

  • A single monolith: a profiler and logs are enough.
  • You need metrics and logs in one solution: Jaeger is only about traces.
  • You lack resources to run trace storage (Elasticsearch, Cassandra) at large volumes.

Key concepts

Trace / Span
A trace consists of spans: operations with a start time, duration and tags.
Sampling
Only part of the traces is kept (for example 1%) so the system is not overloaded.
Service graph
A map of dependencies between services built from spans.
Storage backend
Elasticsearch, Cassandra or Badger for trace storage, in-memory for development only.
OTLP
Jaeger accepts traces over the OpenTelemetry protocol.

Example

Running Jaeger locally in Docker with an OTLP receiver.

Interview questions

  1. 1. What problem does distributed tracing solve?

    In microservices one request passes through many services, and each service's logs alone do not give the whole picture. Tracing collects the full path of the request with timing for each step, so you can see exactly where a delay or error occurs.

  2. 2. What is trace sampling?

    Keeping only a part of the traces to cut costs. There is head sampling (decide at the start of the request) and tail sampling (decide after completion, for example keep only slow or failed ones).

  3. 3. How is Jaeger related to OpenTelemetry?

    OpenTelemetry collects and ships telemetry, while Jaeger stores and displays it. Apps are instrumented through the OTel SDK and Jaeger accepts data over OTLP.

Related technologies

Logos belong to their owners and are used for identification only. Icons: Simple Icons (CC0), Devicon (MIT).

Explore the platform